台北為全國政治、經濟、文化之重心,但因盆地地形之地理條件不佳,極易招致洪災,造成市民生命、財產之損失。近幾年來,世界各國由於都市的發展使降雨集水時間縮短、逕流係數增加,再加上氣候環境變遷,使得原有排水防洪設施之保護標準皆降低,故政府與民間無不投入大量資金與人力努力來改善觀測及警報系統,希望能減少災害發生。一般傳統之預測水文模式或水理模式,會花費相當多的時間做相關參數的調整,若特性改變,模式的重新設定相當繁複,本篇論文收集台北市1998年至2004年19個雨量站之雨量資料與淹水記錄,利用倒傳遞網路模型進行淹水分析,進而預測類似雨量出現時會不會發生淹水,以達預警之功能。本論文運用灰色關聯分析找出對淹水較有影響的因子,以減少倒傳遞網路輸入層節點數量以加快運算時間。本論文除提出模擬結果外,亦詳細說明預測不準的原因。本研究成果將可做為未來驟間降雨是否會造成淹水之參考。
Taipei is the political, economic and cultural center of Taiwan, but the geological structure of Taipei basin is not good as flood disaster causes unexpected loss of life and damage to citizens’ property. In recent years, the intense development of metropolitan cities in countries, worldwide, has shortened the rainfall accumulation time and increased the runoff coefficients of rainfall discharge. Along with the change in global weather and the environment, the original protection standards for drain and flood prevention facilities are comparatively reduced. The government and the private sector invest improvement observing and warning system with diligence and a large amount of fund and manpower to reduce the frequency of flood issues. In this thesis, we collect 19 Taipei regional precipitation stations data from 1998~2004. By utilizing the learning capability of back-propagation neural networks, we can predict the precipitation data and relation of inundation regions. We also apply the grey relational method to extracting the more influential factors for inundation. The extracted factors are then become the inputs of back-propagation network to expedite the learning process. Not only the simulation results are provided, but also a detailed discussion about the false prediction is given. The presented work can be used as a reference for analyzing the occurrence of inundation due to heavy precipitation.